Machine Learning Engineer

ML Engineer - Theseus Group

Theseus Group specializes in automating and enhancing business processes through cutting-edge software, AI, and automation solutions. Our mission is to deliver faster progress with fewer human dependencies, enabling businesses in the United States and Africa to streamline operations and solve complex problems at scale.

*FYI THIS IS NOT A CASUAL JOB

• We work on weekends
• We pay per task, not per hour
• You will be required to use AI, automation, and software to speed up your work. If you take long randomly, you will get paid for the hours the task should've taken with the 3 mentioned
• This job is for those who want mastery and autonomy in work, not money. Your pay will likely be lower than you could get where people are paying less attention

Key Responsibilities

• Architect a hybrid AI pipeline that moves core data scoring off expensive foundational models, utilizing fast embedding models for data retrieval and cross-encoders for precise, cheap scoring.
• Drive massive compute cost reductions by using frontier models to generate training data ("silver labels"), then fine-tuning smaller, cheaper specialized models to handle daily production traffic.
• Deploy dynamic model routing (using ONNX Runtime and NVIDIA Triton).
• Build explicit calibration layers (using techniques like isotonic regression).
• Engineer persistent, stateful memory systems (separating semantic, short-term, and long-term storage) to synthesize patterns, recall user history, and reconstruct personas.

Required Skills

• 3+ years deploying and scaling ML models in production environments.
• Advanced proficiency in Python, PyTorch, and the Hugging Face ecosystem.
• Deep experience with high-performance inference serving (vLLM, ONNX Runtime, NVIDIA Triton).
• Strong applied knowledge of statistical calibration (Scikit-learn) and cost-optimized ML training.
• Demonstrated experience with architectural strategy (PEFT/LoRA).

Preferred Skills

• MLOps for post-deployment monitoring.
• Expertise in Vision-Language Models (VLMs) and multimodal data integration.
• Experience with visual conditioning tools (ControlNet, IP-Adapter) for temporal consistency.
• Industry background in robotics or creative industries (VFX, gaming, spatial computing).